Place recognition for indoor navigation

Abstract

Visual localization related technology has been deeply researched in the recent years, with increasing development in the field of robotics and autonomous vehicle. The project aims to develop an embedded place recognition system to aid navigation in an indoor environment. The FastABLE algorithm was adopted to provide the vision-based methods suitable for mobile devices. The FastABLE algorithm utilizes a set of test and training image sequences to run low level binary sequence extraction using the global binary descriptor and fast matching technique. This meets the requirement of low memory and computational cost to develop a visual navigation system that runs on embedded platforms. The report entails the testing and optimization process of the FastABLE algorithm and the FastABLE android application. The experimental results from the optimized FastABLE android application were subsequently evaluated, achieving average processing time of 1minute 40seconds and average accuracy rate of 48%.